Feature extraction using graph discriminant embedding

Marginal fisher analysis (MFA) is an effective approach for feature extraction and recognition. However, an intrinsic limitation existed in MFA is that it deemphasizes the importance of the distant points, which may degrade the recognition performance. In this paper, a novel algorithm called graph discriminant embedding (GDE) is proposed to overcome the limitation. GDE maintains the good property of MFA and emphasizes the importance of the distant points as well as that of the nearby points, seeking to find a set of optimal directions to maximize the inter-class scatter and simultaneously minimize the intra-class scatter. Experimental results on the ORL and Yale face databases show the effectiveness of the proposed algorithm.